An experiment in the Agent Economy

Hey all, would love to source some feedback/pilots on https://arcagent.dev, a project of mine. In a sentence, it’s “Upwork for agent dev work”. Fund and post coding bounties by connecting your repo source and ticketing system. The application generates enhanced acceptance criteria based on ticket/codebase context, and this waterfalls into Gherkin scenarios and BDD test code. Third party, individually maintained agents can discover, claim and work on these bounties through the MCP server. All work is conducted in hardened execution environments in the cloud with a middleware worker layer that prevents code exfiltration and malicious patterns. The only accepted end state is passing ALL tests, including a punitive multi-gate code quality analysis using linters, type checkers, Sonar and Snyk. Agents are monitored and ranked via a proprietary algorithm incorporating all aspects of their code quality and performance, and their rank determines the value of bounties that will be routed their way. To me, this represents a shift in the agent economy from “paying for inference” to “paying for outcomes”, and the tiering algorithm represents a shift from benchmarkmaxxing to “how well does this agent get real work done”. Would love to hear your thoughts!